Opens in a new tabSkip to content
Agent LighthouseAgent Lighthouse

    Searches the text of every published page. The evidence sources themselves are not in this index — search all of them on the trusted sources page.

    GitHub ↗
    Browse checks and page contents
    operability-safety/contact-form

    Contact/lead form endpoint

    What it checks

    AI agents increasingly handle tasks like “contact this company for a quote” on behalf of users. Without a machine-submittable contact form, agents cannot complete these requests, sending users to competitors who have one. Provide an HTML form or an API endpoint.

    Why it matters

    Exposing contact affordances in server-rendered HTML — tel: and mailto: URIs, a non-JS-dependent contact form, and schema.org ContactPoint on the Organization — increases the rate at which AI agents extract correct contact details and complete contact/booking tasks. Falsifiable: if agents complete contact tasks at the same rate on sites with and without machine-readable contact endpoints, the claim fails.

    Evidence

    Contact and service endpoint discoverability

    The specification layer is solid. ContactPoint is a core schema.org type carrying telephone, email, contactType, areaServed and hoursAvailable, and it is reported on 1M–10M domains. Google explicitly documents supporting contactPoint — ‘The best way for a user to contact your business’ — along with telephone, email and address on Organization. tel: is a ratified IETF Standards Track URI scheme (RFC 3966). The strongest AI-specific argument is derivative rather than direct, and it comes from the rendering gap.

    Vercel and MERJ established that GPTBot and ClaudeBot fetch but never execute JavaScript. A phone number or contact form injected client-side is therefore absent from what those crawlers see, while a server-rendered tel: link is trivially extractable. Google’s generative-AI guide separately notes that ‘Google Business Profiles can help your products and services be visible’ — an off-site contact/service surface with a documented AI-adjacent role.

    Limits

    No AI vendor documents consuming ContactPoint, tel:, or mailto: for task completion. Google’s AI-features page states outright that no special schema.org structured data is needed for AI features, and its Organization page ties contact properties to knowledge-panel display with no AI claim. Architecturally the premise is questionable for the leading agent. ChatGPT agent operates over screenshots of a virtual browser, and interacts through a simulated mouse and keyboard. It engages the rendered form visually rather than parsing machine-readable endpoints, so machine-readable contact metadata is not on its execution path at all.

    BrowserArena’s live-web evaluation finds agent failures concentrated in CAPTCHAs, pop-up banners and direct URL navigation, i.e. in interaction friction rather than in missing contact metadata, so removing modals and keeping forms server-rendered likely matters far more than adding ContactPoint JSON-LD. Keep this informative: the markup is cheap and correct, but no measured agent-completion lift exists.

    Sources